~10% hallucinated sentences: guardrails for lifelogging with LLMs
thejaan · x · 2026-09-24
In a follow-up to his lifelogging thread, Jaan Altosaar details that about 10% of the LLM-generated sentences are made up (e.g. one claimed his dad was recovering from surgery in Estonia), so he listens with vigilance. His guardrails: no memory across sessions, no follow-up questions to avoid reassurance-seeking, a fresh session every 5-10 weeks, and treating it as no substitute for therapy.
More from Apps
- Epic AI MegaJam finals in SF Nov 14-15: teams of 4 plus unlimited AI agents — chrisfirst · 2026-11-12
- Clinical pre/post-surgery video motion data trains one model across neurology, spine, senior care — zawthet · 2026-11-12
- iPad-based gait analysis presented at AAHKS: 110 knees, standing X-ray vs walking alignment — zawthet · 2026-11-12
- Clio Marks One Year Since $1B vLex Deal, Its Fastest-Growing AI Product Signs a Customer Every 12 Minutes — EJWalters · 2026-11-12
- Radical Numerics brings genome language model Omnii-1 to the edge for biodefense — exnx · 2026-11-12
- YC-backed Sintropix launches AI-native ERP that closes books daily and files taxes worldwide — ycombinator · 2026-11-12